-
Notifications
You must be signed in to change notification settings - Fork 41
/
lookup_arrays.jl
963 lines (801 loc) · 33.9 KB
/
lookup_arrays.jl
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
"""
Lookup
Types defining the behaviour of a lookup index, how it is plotted
and how [`Selector`](@ref)s like [`Between`](@ref) work.
A `Lookup` may be [`NoLookup`](@ref) indicating that there are no
lookup values, [`Categorical`](@ref) for ordered or unordered categories,
or a [`Sampled`](@ref) index for [`Points`](@ref) or [`Intervals`](@ref).
"""
abstract type Lookup{T,N} <: AbstractArray{T,N} end
const LookupArray = Lookup
const LookupTuple = Tuple{Lookup,Vararg{Lookup}}
span(lookup::Lookup) = NoSpan()
sampling(lookup::Lookup) = NoSampling()
dims(::Lookup) = nothing
val(l::Lookup) = parent(l)
locus(l::Lookup) = Center()
# Deprecated
index(l::Lookup) = parent(l)
Base.eltype(l::Lookup{T}) where T = T
Base.parent(l::Lookup) = l.data
Base.size(l::Lookup) = size(parent(l))
Base.length(l::Lookup) = length(parent(l))
Base.axes(l::Lookup) = axes(parent(l))
Base.first(l::Lookup) = first(parent(l))
Base.last(l::Lookup) = last(parent(l))
Base.firstindex(l::Lookup) = firstindex(parent(l))
Base.lastindex(l::Lookup) = lastindex(parent(l))
function Base.:(==)(l1::Lookup, l2::Lookup)
basetypeof(l1) == basetypeof(l2) && parent(l1) == parent(l2)
end
ordered_first(l::AbstractArray) = l[ordered_firstindex(l)]
ordered_last(l::AbstractArray) = l[ordered_lastindex(l)]
ordered_firstindex(l::AbstractArray) = firstindex(l)
ordered_firstindex(l::Lookup) = ordered_firstindex(order(l), l)
ordered_firstindex(::ForwardOrdered, l::Lookup) = firstindex(parent(l))
ordered_firstindex(::ReverseOrdered, l::Lookup) = lastindex(parent(l))
ordered_firstindex(::Unordered, l::Lookup) = firstindex(parent(l))
ordered_lastindex(l::AbstractArray) = lastindex(l)
ordered_lastindex(l::Lookup) = ordered_lastindex(order(l), l)
ordered_lastindex(::ForwardOrdered, l::Lookup) = lastindex(parent(l))
ordered_lastindex(::ReverseOrdered, l::Lookup) = firstindex(parent(l))
ordered_lastindex(::Unordered, l::Lookup) = lastindex(parent(l))
function Base.searchsortedfirst(lookup::Lookup, val; lt=<, kw...)
searchsortedfirst(parent(lookup), unwrap(val); order=ordering(order(lookup)), lt=lt, kw...)
end
function Base.searchsortedlast(lookup::Lookup, val; lt=<, kw...)
searchsortedlast(parent(lookup), unwrap(val); order=ordering(order(lookup)), lt=lt, kw...)
end
function Adapt.adapt_structure(to, l::Lookup)
rebuild(l; data=Adapt.adapt(to, parent(l)))
end
"""
AutoLookup <: Lookup
AutoLookup()
AutoLookup(values=AutoValues(); kw...)
Automatic [`Lookup`](@ref), the default lookup. It will be converted automatically
to another [`Lookup`](@ref) when it is possible to detect it from the lookup values.
Keywords will be used in the detected `Lookup` constructor.
"""
struct AutoLookup{T,A<:AbstractVector{T},K} <: Lookup{T,1}
data::A
kw::K
end
AutoLookup(values=AutoValues(); kw...) = AutoLookup(values, kw)
order(lookup::AutoLookup) = hasproperty(lookup.kw, :order) ? lookup.kw.order : AutoOrder()
span(lookup::AutoLookup) = hasproperty(lookup.kw, :span) ? lookup.kw.span : AutoSpan()
sampling(lookup::AutoLookup) = hasproperty(lookup.kw, :sampling) ? lookup.kw.sampling : AutoSampling()
metadata(lookup::AutoLookup) = hasproperty(lookup.kw, :metadata) ? lookup.kw.metadata : NoMetadata()
Base.step(lookup::AutoLookup) = Base.step(parent(lookup))
bounds(lookup::Lookup) = _bounds(order(lookup), lookup)
_bounds(::ForwardOrdered, l::Lookup) = first(l), last(l)
_bounds(::ReverseOrdered, l::Lookup) = last(l), first(l)
_bounds(::Unordered, l::Lookup) = (nothing, nothing)
@noinline Base.step(lookup::T) where T <: Lookup =
error("No step provided by $T. Use a `Sampled` with `Regular`")
"""
Aligned <: Lookup
Abstract supertype for [`Lookup`](@ref)s
where the lookup is aligned with the array axes.
This is by far the most common supertype for `Lookup`.
"""
abstract type Aligned{T,O} <: Lookup{T,1} end
order(lookup::Aligned) = lookup.order
abstract type AbstractNoLookup <: Aligned{Int,Order} end
order(::AbstractNoLookup) = ForwardOrdered()
span(::AbstractNoLookup) = Regular(1)
Base.step(lookup::AbstractNoLookup) = 1
"""
NoLookup <: Lookup
NoLookup()
A [`Lookup`](@ref) that is identical to the array axis.
[`Selector`](@ref)s can't be used on this lookup.
## Example
Defining a `DimArray` without passing lookup values
to the dimensions, it will be assigned `NoLookup`:
```jldoctest NoLookup
using DimensionalData
A = DimArray(rand(3, 3), (X, Y))
Dimensions.lookup(A)
# output
NoLookup, NoLookup
```
Which is identical to:
```jldoctest NoLookup
using .Lookups
A = DimArray(rand(3, 3), (X(NoLookup()), Y(NoLookup())))
Dimensions.lookup(A)
# output
NoLookup, NoLookup
```
"""
struct NoLookup{A<:AbstractVector{Int}} <: AbstractNoLookup
data::A
end
NoLookup() = NoLookup(AutoValues())
rebuild(l::NoLookup; data=parent(l), kw...) = NoLookup(data)
# Used in @d broadcasts
struct Length1NoLookup <: AbstractNoLookup end
Length1NoLookup(::AbstractVector) = Length1NoLookup()
rebuild(l::Length1NoLookup; kw...) = Length1NoLookup()
Base.parent(::Length1NoLookup) = Base.OneTo(1)
"""
AbstractSampled <: Aligned
Abstract supertype for [`Lookup`](@ref)s where the lookup is
aligned with the array, and is independent of other dimensions. [`Sampled`](@ref)
is provided by this package.
`AbstractSampled` must have `order`, `span` and `sampling` fields,
or a `rebuild` method that accepts them as keyword arguments.
"""
abstract type AbstractSampled{T,O<:Order,Sp<:Span,Sa<:Sampling} <: Aligned{T,O} end
span(lookup::AbstractSampled) = lookup.span
sampling(lookup::AbstractSampled) = lookup.sampling
metadata(lookup::AbstractSampled) = lookup.metadata
locus(lookup::AbstractSampled) = locus(sampling(lookup))
Base.step(lookup::AbstractSampled) = step(span(lookup))
function Base.:(==)(l1::AbstractSampled, l2::AbstractSampled)
order(l1) == order(l2) &&
span(l1) == span(l2) &&
sampling(l1) == sampling(l2) &&
parent(l1) == parent(l2)
end
for f in (:getindex, :view, :dotview)
@eval begin
# span may need its step size or bounds updated
@propagate_inbounds function Base.$f(l::AbstractSampled, i::AbstractArray)
i1 = Base.to_indices(l, (i,))[1]
rebuild(l; data=Base.$f(parent(l), i1), span=slicespan(l, i1))
end
end
end
function Adapt.adapt_structure(to, l::AbstractSampled)
rebuild(l; data=Adapt.adapt(to, parent(l)), metadata=NoMetadata(), span=Adapt.adapt(to, span(l)))
end
# bounds
bounds(l::AbstractSampled) = _bounds(order(l), sampling(l), l)
_bounds(order::Order, ::Points, l::AbstractSampled) = _bounds(order, l)
_bounds(::Unordered, ::Intervals, l::AbstractSampled) = (nothing, nothing)
_bounds(::Ordered, sampling::Intervals, l::AbstractSampled) =
_bounds(sampling, span(l), l)
_bounds(::Intervals, span::Irregular, lookup::AbstractSampled) = bounds(span)
_bounds(sampling::Intervals, span::Explicit, lookup::AbstractSampled) =
_bounds(order(lookup), sampling, span, lookup)
_bounds(::ForwardOrdered, ::Intervals, span::Explicit, ::AbstractSampled) =
(val(span)[1, 1], val(span)[2, end])
_bounds(::ReverseOrdered, ::Intervals, span::Explicit, ::AbstractSampled) =
(val(span)[1, end], val(span)[2, 1])
_bounds(::Intervals, span::Regular, lookup::AbstractSampled) =
_bounds(locus(lookup), order(lookup), span, lookup)
_bounds(::Start, ::ForwardOrdered, span, lookup) = first(lookup), last(lookup) + step(span)
_bounds(::Start, ::ReverseOrdered, span, lookup) = last(lookup), first(lookup) - step(span)
_bounds(::Center, ::ForwardOrdered, span, lookup) =
first(lookup) - step(span) / 2, last(lookup) + step(span) / 2
_bounds(::Center, ::ReverseOrdered, span, lookup) =
last(lookup) + step(span) / 2, first(lookup) - step(span) / 2
_bounds(::End, ::ForwardOrdered, span, lookup) = first(lookup) - step(span), last(lookup)
_bounds(::End, ::ReverseOrdered, span, lookup) = last(lookup) + step(span), first(lookup)
const SAMPLED_ARGUMENTS_DOC = """
- `data`: An `AbstractVector` of lookup values, matching the length of the curresponding
array axis.
- `order`: [`Order`](@ref)) indicating the order of the lookup,
[`AutoOrder`](@ref) by default, detected from the order of `data`
to be [`ForwardOrdered`](@ref), [`ReverseOrdered`](@ref) or [`Unordered`](@ref).
These can be provided explicitly if they are known and performance is important.
- `span`: indicates the size of intervals or distance between points, and will be set to
[`Regular`](@ref) for `AbstractRange` and [`Irregular`](@ref) for `AbstractArray`,
unless assigned manually.
- `sampling`: is assigned to [`Points`](@ref), unless set to [`Intervals`](@ref) manually.
Using [`Intervals`](@ref) will change the behaviour of `bounds` and `Selectors`s
to take account for the full size of the interval, rather than the point alone.
- `metadata`: a `Dict` or `Metadata` wrapper that holds any metadata object adding more
information about the array axis - useful for extending DimensionalData for specific
contexts, like geospatial data in Rasters.jl. By default it is `NoMetadata()`.
"""
"""
Sampled <: AbstractSampled
Sampled(data::AbstractVector, order::Order, span::Span, sampling::Sampling, metadata)
Sampled(data=AutoValues(); order=AutoOrder(), span=AutoSpan(), sampling=Points(), metadata=NoMetadata())
A concrete implementation of the [`Lookup`](@ref)
[`AbstractSampled`](@ref). It can be used to represent
[`Points`](@ref) or [`Intervals`](@ref).
`Sampled` is capable of representing gridded data from a wide range of sources,
allowing correct `bounds` and [`Selector`](@ref)s for points or intervals of
regular, irregular, forward and reverse lookups.
On `AbstractDimArray` construction, `Sampled` lookup is assigned for all lookups of
`AbstractRange` not assigned to [`Categorical`](@ref).
## Arguments
$SAMPLED_ARGUMENTS_DOC
## Example
Create an array with `Interval` sampling, and `Regular` span for a vector with known spacing.
We set the [`locus`](@ref) of the `Intervals` to `Start` specifying
that the lookup values are for the locus at the start of each interval.
```jldoctest Sampled
using DimensionalData, DimensionalData.Lookups
x = X(Sampled(100:-20:10; sampling=Intervals(Start())))
y = Y(Sampled([1, 4, 7, 10]; span=Regular(3), sampling=Intervals(Start())))
A = ones(x, y)
# output
╭─────────────────────────╮
│ 5×4 DimArray{Float64,2} │
├─────────────────────────┴────────────────────────────────────────── dims ┐
↓ X Sampled{Int64} 100:-20:20 ReverseOrdered Regular Intervals{Start},
→ Y Sampled{Int64} [1, 4, 7, 10] ForwardOrdered Regular Intervals{Start}
└──────────────────────────────────────────────────────────────────────────┘
↓ → 1 4 7 10
100 1.0 1.0 1.0 1.0
80 1.0 1.0 1.0 1.0
60 1.0 1.0 1.0 1.0
40 1.0 1.0 1.0 1.0
20 1.0 1.0 1.0 1.0
```
"""
struct Sampled{T,A<:AbstractVector{T},O,Sp,Sa,M} <: AbstractSampled{T,O,Sp,Sa}
data::A
order::O
span::Sp
sampling::Sa
metadata::M
end
function Sampled(data=AutoValues();
order=AutoOrder(), span=AutoSpan(),
sampling=AutoSampling(), metadata=NoMetadata()
)
Sampled(data, order, span, sampling, metadata)
end
function rebuild(l::Sampled;
data=parent(l), order=order(l), span=span(l), sampling=sampling(l), metadata=metadata(l), kw...
)
Sampled(data, order, span, sampling, metadata)
end
# These are used to specialise dispatch:
# When Cycling, we need to modify any `Selector`. After that
# we switch to `NotCycling` and use `AbstractSampled` fallbacks.
# We could switch to `Sampled` at that point, but its less extensible.
abstract type CycleStatus end
struct Cycling <: CycleStatus end
struct NotCycling <: CycleStatus end
"""
AbstractCyclic <: AbstractSampled
An abstract supertype for cyclic lookups.
These are `AbstractSampled` lookups that are cyclic for `Selectors`.
"""
abstract type AbstractCyclic{X,T,O,Sp,Sa} <: AbstractSampled{T,O,Sp,Sa} end
cycle(l::AbstractCyclic) = l.cycle
cycle_status(l::AbstractCyclic) = l.cycle_status
bounds(l::AbstractCyclic{<:Any,T}) where T = (typemin(T), typemax(T))
# Indexing with `AbstractArray` must rebuild the lookup as
# `Sampled` as we no longer have the whole cycle.
for f in (:getindex, :view, :dotview)
@eval @propagate_inbounds Base.$f(l::AbstractCyclic, i::AbstractArray) =
Sampled(rebuild(l; data=Base.$f(parent(l), i)))
end
no_cycling(l::AbstractCyclic) = rebuild(l; cycle_status=NotCycling())
function cycle_val(l::AbstractCyclic, val)
cycle_start = ordered_first(l)
# This formulation is necessary for dates
ncycles = (val - cycle_start) ÷ (cycle_start + cycle(l) - cycle_start)
res = val - ncycles * cycle(l)
# Catch precision errors
if (cycle_start + (ncycles + 1) * cycle(l)) <= val
i = 1
while i < 10000
if (cycle_start + (ncycles + i) * cycle(l)) > val
return val - (ncycles + i - 1) * cycle(l)
end
i += 1
end
elseif res < cycle_start
i = 1
while i < 10000
res = val - (ncycles - i + 1) * cycle(l)
res >= cycle_start && return res
i += 1
end
else
return res
end
error("`Cyclic` lookup too innacurate, value not found")
end
"""
Cyclic <: AbstractCyclic
Cyclic(data; order=AutoOrder(), span=AutoSpan(), sampling=Points(), metadata=NoMetadata(), cycle)
A `Cyclic` lookup is similar to `Sampled` but out of range `Selectors` [`At`](@ref),
[`Near`](@ref), [`Contains`](@ref) will cycle the values to `typemin` or `typemax`
over the length of `cycle`. [`Where`](@ref) and `..` work as for [`Sampled`](@ref).
This is useful when we are using mean annual datasets over a real time-span,
or for wrapping longitudes so that `-360` and `360` are the same.
## Arguments
$SAMPLED_ARGUMENTS_DOC
- `cycle`: the length of the cycle. This does not have to exactly match the data,
the `step` size is `Week(1)` the cycle can be `Years(1)`.
## Notes
1. If you use dates and e.g. cycle over a `Year`, every year will have the
number and spacing of `Week`s and `Day`s as the cycle year. Using `At` may not be reliable
in terms of exact dates, as it will be applied to the specified date plus or minus `n` years.
2. Indexing into a `Cycled` with any `AbstractArray` or `AbstractRange` will return
a [`Sampled`](@ref) as the full cycle is likely no longer available.
3. `..` or `Between` selectors do not work in a cycled way: they work as for [`Sampled`](@ref).
This may change in future to return cycled values, but there are problems with this, such as
leap years breaking correct date cycling of a single year. If you actually need this behaviour,
please make a GitHub issue.
"""
struct Cyclic{X,T,A<:AbstractVector{T},O,Sp,Sa,M,C} <: AbstractCyclic{X,T,O,Sp,Sa}
data::A
order::O
span::Sp
sampling::Sa
metadata::M
cycle::C
cycle_status::X
function Cyclic(
data::A, order::O, span::Sp, sampling::Sa, metadata::M, cycle::C, cycle_status::X
) where {A<:AbstractVector{T},O,Sp,Sa,M,C,X} where T
_check_ordered_cyclic(order)
new{X,T,A,O,Sp,Sa,M,C}(data, order, span, sampling, metadata, cycle, cycle_status)
end
end
function Cyclic(data=AutoValues();
order=AutoOrder(), span=AutoSpan(),
sampling=AutoSampling(), metadata=NoMetadata(),
cycle, # Mandatory keyword, there are too many possible bugs with auto detection
)
cycle_status = Cycling()
Cyclic(data, order, span, sampling, metadata, cycle, cycle_status)
end
_check_ordered_cyclic(::AutoOrder) = nothing
_check_ordered_cyclic(::Ordered) = nothing
_check_ordered_cyclic(::Unordered) = throw(ArgumentError("Cyclic lookups must be `Ordered`"))
function rebuild(l::Cyclic;
data=parent(l), order=order(l), span=span(l), sampling=sampling(l), metadata=metadata(l),
cycle=cycle(l), cycle_status=cycle_status(l), kw...
)
Cyclic(data, order, span, sampling, metadata, cycle, cycle_status)
end
"""
AbstractCategorical <: Aligned
[`Lookup`](@ref)s where the values are categories.
[`Categorical`](@ref) is the provided concrete implementation.
But this can easily be extended, all methods are defined for `AbstractCategorical`.
All `AbstractCategorical` must provide a `rebuild`
method with `data`, `order` and `metadata` keyword arguments.
"""
abstract type AbstractCategorical{T,O} <: Aligned{T,O} end
order(lookup::AbstractCategorical) = lookup.order
metadata(lookup::AbstractCategorical) = lookup.metadata
const CategoricalEltypes = Union{AbstractChar,Symbol,AbstractString}
function Adapt.adapt_structure(to, l::AbstractCategorical)
rebuild(l; data=Adapt.adapt(to, parent(l)), metadata=NoMetadata())
end
"""
Categorical <: AbstractCategorical
Categorical(o::Order)
Categorical(; order=Unordered())
A [`Lookup`](@ref) where the values are categories.
This will be automatically assigned if the lookup contains `AbstractString`,
`Symbol` or `Char`. Otherwise it can be assigned manually.
[`Order`](@ref) will be determined automatically where possible.
## Arguments
- `data`: An `AbstractVector` matching the length of the corresponding
array axis.
- `order`: [`Order`](@ref)) indicating the order of the lookup,
[`AutoOrder`](@ref) by default, detected from the order of `data`
to be `ForwardOrdered`, `ReverseOrdered` or `Unordered`.
Can be provided if this is known and performance is important.
- `metadata`: a `Dict` or `Metadata` wrapper that holds any metadata object adding more
information about the array axis - useful for extending DimensionalData for specific
contexts, like geospatial data in Rasters.jl. By default it is `NoMetadata()`.
## Example
Create an array with [`Interval`] sampling.
```jldoctest Categorical
using DimensionalData
ds = X(["one", "two", "three"]), Y([:a, :b, :c, :d])
A = DimArray(rand(3, 4), ds)
Dimensions.lookup(A)
# output
Categorical{String} ["one", "two", "three"] Unordered,
Categorical{Symbol} [:a, :b, :c, :d] ForwardOrdered
```
"""
struct Categorical{T,A<:AbstractVector{T},O<:Order,M} <: AbstractCategorical{T,O}
data::A
order::O
metadata::M
end
function Categorical(data=AutoValues(); order=AutoOrder(), metadata=NoMetadata())
Categorical(data, order, metadata)
end
function rebuild(l::Categorical;
data=parent(l), order=order(l), metadata=metadata(l), kw...
)
Categorical(data, order, metadata)
end
function Base.:(==)(l1::AbstractCategorical, l2::AbstractCategorical)
order(l1) == order(l2) && parent(l1) == parent(l2)
end
"""
Unaligned <: Lookup
Abstract supertype for [`Lookup`](@ref) where the lookup is not aligned to the grid.
Indexing an [`Unaligned`](@ref) with [`Selector`](@ref)s must provide all
other [`Unaligned`](@ref) dimensions.
"""
abstract type Unaligned{T,N} <: Lookup{T,N} end
"""
Transformed <: Unaligned
Transformed(f, dim::Dimension; metadata=NoMetadata())
[`Lookup`](@ref) that uses an affine transformation to convert
dimensions from `dims(lookup)` to `dims(array)`. This can be useful
when the dimensions are e.g. rotated from a more commonly used axis.
Any function can be used to do the transformation, but transformations
from CoordinateTransformations.jl may be useful.
## Arguments
- `f`: transformation function
- `dim`: a dimension to transform to.
## Keyword Arguments
- `metadata`:
## Example
```jldoctest
using DimensionalData, DimensionalData.Lookups, CoordinateTransformations
m = LinearMap([0.5 0.0; 0.0 0.5])
A = [1 2 3 4
5 6 7 8
9 10 11 12];
da = DimArray(A, (X(Transformed(m)), Y(Transformed(m))))
da[X(At(6.0)), Y(At(2.0))]
# output
9
```
"""
struct Transformed{T,A<:AbstractVector{T},F,D,M} <: Unaligned{T,1}
data::A
f::F
dim::D
metadata::M
end
function Transformed(f; metadata=NoMetadata())
Transformed(AutoValues(), f, AutoDim(), metadata)
end
function Transformed(f, data::AbstractArray; metadata=NoMetadata())
Transformed(data, f, AutoDim(), metadata)
end
function rebuild(l::Transformed;
data=parent(l), f=transformfunc(l), dim=dim(l), metadata=metadata(l)
)
Transformed(data, f, dim, metadata)
end
dim(lookup::Transformed) = lookup.dim
transformfunc(lookup::Transformed) = lookup.f
Base.:(==)(l1::Transformed, l2::Transformed) = typeof(l1) == typeof(l2) && f(l1) == f(l2)
# TODO Transformed bounds
# Shared methods
intervalbounds(l::Lookup, args...) = _intervalbounds_no_interval_error()
intervalbounds(l::AbstractSampled, args...) = intervalbounds(span(l), sampling(l), l, args...)
intervalbounds(span::Span, ::Points, ls::Lookup) = map(l -> (l, l), ls)
intervalbounds(span::Span, ::Points, ls::Lookup, i::Int) = ls[i], ls[i]
intervalbounds(span::Span, sampling::Intervals, l::Lookup, i::Int) =
intervalbounds(order(l), locus(sampling), span, l, i)
function intervalbounds(order::ForwardOrdered, locus::Start, span::Span, l::Lookup, i::Int)
if i == lastindex(l)
(l[i], bounds(l)[2])
else
(l[i], l[i+1])
end
end
function intervalbounds(order::ForwardOrdered, locus::End, span::Span, l::Lookup, i::Int)
if i == firstindex(l)
(bounds(l)[1], l[i])
else
(l[i-1], l[i])
end
end
function intervalbounds(order::ReverseOrdered, locus::Start, span::Span, l::Lookup, i::Int)
if i == firstindex(l)
(l[i], bounds(l)[2])
else
(l[i], l[i-1])
end
end
function intervalbounds(order::ReverseOrdered, locus::End, span::Span, l::Lookup, i::Int)
if i == lastindex(l)
(bounds(l)[1], l[i])
else
(l[i+1], l[i])
end
end
# Regular Center
function intervalbounds(order::Ordered, locus::Center, span::Regular, l::Lookup, i::Int)
halfstep = step(span) / 2
x = l[i]
bounds = (x - halfstep, x + halfstep)
return _maybeflipbounds(order, bounds)
end
# Irregular Center
function intervalbounds(order::ForwardOrdered, locus::Center, span::Irregular, l::Lookup, i::Int)
x = l[i]
low = i == firstindex(l) ? bounds(l)[1] : x + (l[i - 1] - x) / 2
high = i == lastindex(l) ? bounds(l)[2] : x + (l[i + 1] - x) / 2
return (low, high)
end
function intervalbounds(order::ReverseOrdered, locus::Center, span::Irregular, l::Lookup, i::Int)
x = l[i]
low = i == firstindex(l) ? bounds(l)[2] : x + (l[i - 1] - x) / 2
high = i == lastindex(l) ? bounds(l)[1] : x + (l[i + 1] - x) / 2
return (low, high)
end
function intervalbounds(span::Span, sampling::Intervals, l::Lookup)
map(axes(l, 1)) do i
intervalbounds(span, sampling, l, i)
end
end
# Explicit
function intervalbounds(span::Explicit, ::Intervals, l::Lookup, i::Int)
return (l[1, i], l[2, i])
end
# We just reinterpret the bounds matrix rather than allocating
function intervalbounds(span::Explicit, ::Intervals, l::Lookup)
m = val(span)
T = eltype(m)
return reinterpret(reshape, Tuple{T,T}, m)
end
_intervalbounds_no_interval_error() = error("Lookup does not have Intervals, `intervalbounds` cannot be applied")
# slicespan should only be called after `to_indices` has simplified indices
slicespan(l::Lookup, i::Colon) = span(l)
slicespan(l::Lookup, i) = _slicespan(span(l), l, i)
_slicespan(span::Regular, l::Lookup, i::Union{AbstractRange,CartesianIndices}) = Regular(step(l) * step(i))
_slicespan(span::Regular, l::Lookup, i::AbstractArray) = _slicespan(Irregular(bounds(l)), l, i)
_slicespan(span::Explicit, l::Lookup, i::AbstractArray) = Explicit(val(span)[:, i])
_slicespan(span::Irregular, l::Lookup, i::AbstractArray) =
_slicespan(sampling(l), span, l, i)
function _slicespan(span::Irregular, l::Lookup, i::Base.LogicalIndex)
i1 = length(i) == 0 ? (1:0) : ((findfirst(i.mask)::Int):(findlast(i.mask)::Int))
_slicespan(sampling(l), span, l, i1)
end
function _slicespan(span::Irregular, l::Lookup, i::InvertedIndices.InvertedIndexIterator)
i1 = collect(i) # We could do something more efficient here, but I'm not sure what
_slicespan(sampling(l), span, l, i1)
end
_slicespan(::Points, span::Irregular, l::Lookup, i::AbstractArray) =
Irregular(nothing, nothing)
_slicespan(::Intervals, span::Irregular, l::Lookup, i::AbstractArray) =
Irregular(_slicebounds(span, l, i))
function _slicebounds(span::Irregular, l::Lookup, i::AbstractArray)
length(i) == 0 && return (nothing, nothing)
_slicebounds(locus(l), span, l, i)
end
function _slicebounds(locus::Start, span::Irregular, l::Lookup, i::AbstractArray)
fi, la = first(i), last(i)
if isforward(l)
l[fi], la >= lastindex(l) ? bounds(l)[2] : l[la + 1]
else
l[la], fi <= firstindex(l) ? bounds(l)[2] : l[fi - 1]
end
end
function _slicebounds(locus::End, span::Irregular, l::Lookup, i::AbstractArray)
fi, la = first(i), last(i)
if isforward(l)
fi <= firstindex(l) ? bounds(l)[1] : l[fi - 1], l[la]
else
la >= lastindex(l) ? bounds(l)[1] : l[la + 1], l[fi]
end
end
function _slicebounds(locus::Center, span::Irregular, l::Lookup, i::AbstractArray)
fi, la = first(i), last(i)
a, b = if isforward(l)
fi <= firstindex(l) ? bounds(l)[1] : (l[fi - 1] + l[fi]) / 2,
la >= lastindex(l) ? bounds(l)[2] : (l[la + 1] + l[la]) / 2
else
la >= lastindex(l) ? bounds(l)[1] : (l[la + 1] + l[la]) / 2,
fi <= firstindex(l) ? bounds(l)[2] : (l[fi - 1] + l[fi]) / 2
end
return a, b
end
# Have to special-case date/time so we work with seconds and add to the original
function _slicebounds(locus::Center, span::Irregular, l::Lookup{T}, i::AbstractArray) where T<:Dates.AbstractTime
op = T === Date ? div : /
frst = if first(i) <= firstindex(l)
_maybeflipbounds(l, bounds(l))[1]
else
if isrev(order(l))
op(l[first(i)] - l[first(i) - 1], 2) + l[first(i) - 1]
else
op(l[first(i) - 1] - l[first(i)], 2) + l[first(i)]
end
end
lst = if last(i) >= lastindex(l)
_maybeflipbounds(l, bounds(l))[2]
else
if isrev(order(l))
op(l[last(i)] - l[last(i) + 1], 2) + l[last(i) + 1]
else
op(l[last(i) + 1] - l[last(i)], 2) + l[last(i)]
end
end
return (frst, lst)
end
# reducing methods
@inline reducelookup(lookup::NoLookup) = NoLookup(OneTo(1))
# TODO what should this do?
@inline reducelookup(lookup::Unaligned) = NoLookup(OneTo(1))
# Categories are combined.
@inline reducelookup(lookup::Categorical{<:AbstractString}) =
rebuild(lookup; data=["combined"])
@inline reducelookup(lookup::Categorical) = rebuild(lookup; data=[:combined])
# Sampled is resampled
@inline reducelookup(lookup::AbstractSampled) = _reducelookup(span(lookup), lookup)
@inline _reducelookup(::Irregular, lookup::AbstractSampled) = begin
rebuild(lookup; data=_reducevalues(lookup), order=ForwardOrdered())
end
@inline _reducelookup(span::Regular, lookup::AbstractSampled) = begin
newstep = step(span) * length(lookup)
newvalues = _reducevalues(lookup, newstep)
# Make sure the step type matches the new eltype
newstep = convert(promote_type(eltype(newvalues), typeof(newstep)), newstep)
newspan = Regular(newstep)
rebuild(lookup; data=newvalues, order=ForwardOrdered(), span=newspan)
end
@inline _reducelookup(
span::Regular{<:Dates.CompoundPeriod}, lookup::AbstractSampled
) = begin
newstep = Dates.CompoundPeriod(step(span).periods .* length(lookup))
# We don't pass the step here - the range doesn't work with CompoundPeriod
newvalues = _reducevalues(lookup)
# Make sure the step type matches the new eltype
newspan = Regular(newstep)
rebuild(lookup; data=newvalues, order=ForwardOrdered(), span=newspan)
end
@inline _reducelookup(span::Explicit, lookup::AbstractSampled) = begin
bnds = val(span)
newstep = bnds[2] - bnds[1]
newvalues = _reducevalues(lookup, newstep)
# Make sure the step type matches the new eltype
newstep = convert(promote_type(eltype(newvalues), typeof(newstep)), newstep)
newspan = Explicit(reshape([bnds[1, 1]; bnds[2, end]], 2, 1))
newlookup = rebuild(lookup; data=newvalues, order=ForwardOrdered(), span=newspan)
end
# Get the lookup value at the reduced locus.
# This is the start, center or end point of the whole lookup.
@inline _reducevalues(lookup::Lookup, step=nothing) = _reducevalues(locus(lookup), lookup, step)
@inline _reducevalues(locus::Start, lookup::Lookup, step) = _mayberange(first(lookup), step)
@inline _reducevalues(locus::End, lookup::Lookup, step) = _mayberange(last(lookup), step)
@inline _reducevalues(locus::Center, lookup::Lookup, step) = begin
values = parent(lookup)
len = length(values)
newval = centerval(values, len)
_mayberange(newval, step)
end
# Ranges with a known step always return a range
_mayberange(x, step) = x:step:x
# Arrays return a vector
_mayberange(x, step::Nothing) = [x]
@inline centerval(values::AbstractArray{<:Number}, len) = (first(values) + last(values)) / 2
@inline function centerval(values::AbstractArray{<:DateTime}, len)
f = first(values)
l = last(values)
if f <= l
return (l - f) / 2 + first(values)
else
return (f - l) / 2 + last(values)
end
end
@inline centerval(values::AbstractArray, len) = values[len ÷ 2 + 1]
ordering(::ForwardOrdered) = Base.Order.ForwardOrdering()
ordering(::ReverseOrdered) = Base.Order.ReverseOrdering()
# Promotion
# General case
promote_first(x) = x
promote_first(x1, x2, xs...) =
convert(promote_type(typeof(x1), typeof(x2), map(typeof, xs)...), x1)
# Fallback NoLookup if not identical type
promote_first(l1::Lookup) = l1
promote_first(l1::L, ls::L...) where L<:Lookup = rebuild(l1; metadata=NoMetadata)
function promote_first(l1::L, ls::Lookup...) where {L<:Lookup}
ls = _remove(Length1NoLookup, l1, ls...)
if length(ls) > 1
l1, ls... = ls
else
return first(ls)
end
if all(map(l -> typeof(l) == L, ls))
if length(ls) > 0
rebuild(l1; metadata=NoMetadata())
else
l1 # Keep metadata if there is only one lookup
end
else
NoLookup(Base.OneTo(length(l1)))
end
end
# Categorical lookups
promote_first(l1::AbstractCategorical) = l1
promote_first(l1::C, ls::C...) where C<:AbstractCategorical = l1
promote_first(l1::C, ::C, ::C...) where C<:AbstractCategorical = rebuild(l1; metadata=NoMetadata())
function promote_first(l1::AbstractCategorical, l2::AbstractCategorical, ls::AbstractCategorical...)
ls = (l2, ls...)
o = all(map(l -> order(l) == order(l1), ls)) ? order(l1) : Unordered()
data = promote_first(parent(l1), map(parent, ls)...)
# Check we have all the same type of AbstractCategorical
if all(map(l -> basetypeof(l) == basetypeof(l1), ls))
return rebuild(l1; data, order=o, metadata=NoMetadata())
else # Otherwise fall back to Categorical
return Categorical(data; order=o, metadata=NoMetadata())
end
end
promote_first(l1::AbstractSampled) = l1
promote_first(l1::S, ::S, ::S...) where S<:AbstractSampled = l1
function promote_first(l1::AbstractSampled, l2::AbstractSampled, ls::AbstractSampled...)
ls = (l2, ls...)
# We cant always convert explicit to something else
if any(map(isexplicit, (l1, ls...))) && !all(isexplicit, (l1, ls...))
return NoLookup(Base.OneTo(length(l1)))
end
data = promote_first(parent(l1), map(parent, ls)...)
sa = promote_first(sampling(l1), map(sampling, ls)...)
kw = (;
order=promote_first(order(l1), map(order, ls)...),
sampling=sa,
span=promote_first(l1, sa, span(l1), map(span, ls)...),
metadata=NoMetadata(),
)
# Check we have all the same type of AbstractSampled
if all(map(l -> basetypeof(l) == basetypeof(l1), ls))
return rebuild(l1; data, kw...)
else # Otherwise fall back to Sampled
return Sampled(data; kw...)
end
end
# Order
# Only matching Order remain the same
promote_first(::O, ::O...) where O<:Order = O()
# Everthing else is Unordered
promote_first(::Order, ::Order...) = Unordered()
# Sampling
# Only matching locus Intervals remain Intervals
promote_first(i1::I, ::I...) where I<:Intervals = first(i1)
# Any other mix is Points
promote_first(::Sampling, ::Sampling...) = Points()
# Span
# Regular remains regular, eltype is promoted
function promote_first(::Lookup, ::Sampling, s::Regular, ss::Regular...)
T = promote_type(typeof(val(s)), map(typeof ∘ val, ss)...)
Regular(convert(T, val(s)))
end
# # Matching irregular is returns
promote_first(::Lookup, ::Sampling, a::T, b::T...) where T<:Irregular = a
# # Number and DateTime are promoted
for E in (Base.Number, Dates.AbstractTime)
@eval function promote_first(::Lookup,
s::Irregular{Tuple{<:$E,<:$E}}, ss::Irregular{Tuple{<:$E,<:$E}}...
)
T = promote_type(maps(s -> promote_type(typeof(val(s)[1]), typeof(val(s)[2])), (s, ss...))...)
return Irregular(convert(T, val(a)[1]), convert(T, val(a)[2]))
end
end
# Explicit promotes its matrix
promote_first(::Lookup, ::Sampling, s1::Explicit, ss::Explicit...) =
Explicit(promote_first(val(s1), map(val, spans)...))
# Mixed Regular/Irregular always become Irregular
promote_first(l::Lookup, sampling::Sampling, ::Union{Regular,Irregular}, ::Union{Regular,Irregular}...) =
_irregular(sampling, l)
_irregular(::Points, l) = Irregular(nothing, nothing)
_irregular(::Intervals, l) = Irregular(bounds(l))
# Data
promote_first(a1::A) where A<:AbstractArray = a1
promote_first(a1::A, ::A, ::A...) where A<:AbstractArray = a1
promote_first(a1::AbstractArray{<:AbstractString}, as::AbstractArray{<:AbstractString}...) = String.(a1)
function promote_first(a1::AbstractArray, as::AbstractArray...)
T = promote_type(eltype(a1), map(eltype, as)...)
C = if a1 isa AbstractRange && all(map(a -> a isa AbstractRange, as))
if a1 isa AbstractUnitRange && all(map(a -> a isa AbstractUnitRange, as))
UnitRange
elseif a1 isa OrdinalRange && all(map(a -> a isa OrdinalRange, as))
S = promote_type(typeof(step(a1)), map(typeof ∘ step, as)...)
StepRange{T,S}
elseif a1 isa LinRange || any(map(a -> a isa LinRange, as))
LinRange{T}
else
StepRangeLen{T}
end
else
Vector{T}
end
return convert(C, a1)
end